A neural network controller with PID compensation for trajectory tracking of robotic manipulators

A simple and effective method for trajectory tracking of robotic manipulator is proposed. The first step is to employ a neural network to learn the characteristics of the inverse dynamics of the robotic manipulator in an off-line manner. Then the neural network is placed in series with the robotic manipulator for on-line operation and the desired trajectory is fed into the neural network to obtain the corresponding torques. If all the characteristics of the inverse dynamics are learned by the neural network, then the corresponding torques will generate the desired trajectory. However, considering that the input patterns of the neural network may not be rich enough to excite all modes of the system, a classical PID controller is added for compensation. This controller is placed at the input node of the robotic manipulator to compensate the characteristics that have not been learned by the neural network. In this control architecture, it will be shown that the robotic manipulator can track precisely both the trained and untrained trajectories. Simulation examples are included to illustrate the validity of the proposed method.